Incident sheet · /enterprise-ai
- System
- Enterprise AI
- Filed
- Read
- 4 min · 758 words
- Severity
- Routine
Microsoft launches MAI Code 2 models to cut reliance on OpenAI
MAI Code 2 scored 63.2% average on four Copilot benchmarks against 64.1% for GPT-6.1 Sol. Promotional token pricing runs until May 17, 2027, then doubles.

Event log
- MAI Code 2 averaged 63.2% across four benchmarks versus 64.1% for GPT-6.1 Sol and 65.3% for Claude Opus 5.5.
- Average cost per task: $0.309 for MAI Code 2 against $6.306 for Claude Opus 5.5, at 50% promotional pricing.
- Promotional pricing of $0.50 per million input tokens for MAI Code 2 ends May 17, 2027, then doubles.
- MAI Code 2 Flash averaged 61%, ahead of GPT-6 Luna at 55.1% and Claude Haiku 5.5 at 53.9%.
- Jain: a 60% model-cost cut yields only 6% overall savings if models account for 10% of development costs.
Microsoft has launched two proprietary AI coding models, MAI Code 2 and MAI Code 2 Flash, with an average benchmark pass rate of 63.2% against 64.1% for OpenAI's GPT-6.1 Sol and 65.3% for Anthropic's Claude Opus 5.5. The company unveiled the models at Ignite 2026. The move reduces Microsoft's dependence on the external partners that fueled its generative AI push.
What does each model target?
MAI Code 2 serves common enterprise coding tasks. Microsoft lists deep debugging, changes across multiple files, repository-wide reasoning and user interface development as its use cases. MAI Code 2 Flash targets high-volume work: code explanations, targeted edits, routine fixes and rapid iterations. Microsoft said in its press statement that the models are explicitly not yet tuned for harder problems such as fundamental science research and long-horizon agentic work.
How did the models score?
Microsoft ran comparisons across four widely used benchmarks (standardized coding test suites) in GitHub Copilot: DeepSWE 1.1, Terminal-Bench 2.1, SWE-Atlas and FrontierCode v1.1.
MAI Code 2 scores versus GPT-6.1 Sol and Claude Opus 5.5:
- DeepSWE 1.1: 73.2% versus 74.6% and 69%
- Terminal-Bench 2.1: 84.6% versus 85% and 78.7%
- SWE-Atlas: 43.1% versus 47.4% and 57.5%
- FrontierCode v1.1: 52% versus 49.4% and 56%
Microsoft compared MAI Code 2 Flash with GPT-6 Luna and Claude Haiku 5.5 across three benchmarks at maximum reasoning. The Flash model averaged 61%, ahead of GPT-6 Luna at 55.1% and Claude Haiku 5.5 at 53.9%. Individual scores: 59.9% on DeepSWE 1.1, 78.3% on Terminal-Bench 2.1 and 44.8% on SWE-Atlas. GPT-6 Luna scored 59.6%, 69.3% and 36.4%; Claude Haiku 5.5 scored 46%, 73.4% and 42.1%.
What does it cost?
Microsoft claimed an average cost per task of $0.309 for MAI Code 2, against $0.761 for GPT-6.1 Sol and $6.306 for Claude Opus 5.5. MAI Code 2 Flash averaged $0.052 per task, versus $0.080 for GPT-6 Luna and $0.319 for Claude Haiku 5.5.
The comparison rests on 50% promotional pricing. According to GitHub's pricing page, MAI Code 2 costs $0.50 per million input tokens, $0.025 per million cached input tokens and $2.50 per million output tokens until May 17, 2027. Prices then double to $1.00, $0.05 and $5.00 respectively. MAI Code 2 Flash carries promotional rates of $0.05 per million input tokens, $0.005 per million cached input tokens and $0.25 per million output tokens, rising later to $0.10, $0.01 and $0.50.
"Microsoft's own savings are measured at a 50% promotional price, so the gap narrows sharply once the promotion ends. And the cost it compares is the spend of a benchmark run, not a published price per token," said Sanchit Vir Gogia, chief analyst at Greyhound Research.
Why is Microsoft building its own models?
Pareekh Jain, CEO at EIIRTrend and Pareekh Consulting, said the launch reduces Microsoft's dependence on OpenAI and Anthropic. "Microsoft also has a major advantage through GitHub, Visual Studio and Azure, which are already used by millions of developers. Its immediate goal is to strengthen its developer ecosystem and reduce AI costs. So it doesn't have to pay OpenAI and Anthropic every time someone uses GitHub Copilot. Microsoft is protecting its profits and locking developers into its own tools," he said.
What should CIOs watch?
Model costs cover only part of total expense. "Enterprises must also pay for integration, testing, security, human reviews and fixing AI-generated errors," said Jain. He calculated that a 60% reduction in model costs delivers only 6% overall savings if model usage accounts for 10% of total development costs.
Gogia advised technology leaders to test the models on their own code, through their own tools, at the effort setting intended for production, then track cost per merged code change, change failure rates and security findings against a baseline from teams not using AI coding models. He cited Google's DORA research, which has associated wider AI adoption with weaker delivery stability.
Security adds another concern. Microsoft's model cards rate both models highly on offensive cyber capabilities. Gogia said pilot programmes should assess not just the quality of AI-generated code but also its security and reliability when integrated into production systems.
via infoworld.com/feed/ (Source)
More from Rebecca Stone
Show full bio
Senior reporter covering money, funding and market shifts around enterprise information technology and the leaders who run at CIO Monitor.
195 articles
The newsroom digest
One email a day with the stories that matter. No noise.
Cross-references
- Microsoft launches MAI-Code-2 at half GPT-6.1’s token price
- Microsoft Fires Back at OpenAI and Anthropic With MAI Code 2
- Microsoft's MAI models chase cheap tokens, not bragging rights
- OpenAI Ships GPT-6.1 Terra at Half the Price of Its Predecessor
- OpenAI Cuts Token Prices 20% With GPT-6.2 Astra and Sol Ahead of Anthropic IPO